DocuPurse / src /classifier.py
Zayeemk's picture
Initial commit: DocuParse AI complete implementation
26b8deb
Raw History Blame Contribute Delete
2.02 kB
"""Document type classifier using keyword heuristics."""
from __future__ import annotations
import re
from typing import Tuple
_INVOICE_KEYWORDS = [
r"\binvoice\b", r"\binv[-\s]?\d", r"\bbill\s+to\b", r"\bdue\s+date\b",
r"\btax\s+invoice\b", r"\bgstin\b", r"\bremit\s+to\b", r"\bpayment\s+due\b",
r"\binvoice\s+number\b", r"\binvoice\s+no\b", r"\binvoice\s+date\b",
]
_RECEIPT_KEYWORDS = [
r"\breceipt\b", r"\bcashier\b", r"\bamount\s+paid\b", r"\bpayment\s+received\b",
r"\bthank\s+you\s+for\b", r"\bchange\b", r"\bcash\s+tender\b", r"\breceived\s+from\b",
r"\bpos\s+terminal\b", r"\btransaction\s+id\b",
]
_PO_KEYWORDS = [
r"\bpurchase\s+order\b", r"\bpo\s+number\b", r"\bpo\s+no\b",
r"\bship\s+to\b", r"\border\s+date\b", r"\bdelivery\s+date\b",
r"\bvendor\b", r"\bbuyer\b", r"\bprocurement\b", r"\brequisition\b",
]
def _count_matches(text: str, patterns: list) -> int:
count = 0
t = text.lower()
for pattern in patterns:
if re.search(pattern, t):
count += 1
return count
def classify_document(text: str) -> Tuple[str, float]:
"""
Classify document type from extracted text.
Returns (document_type, confidence) where confidence is 0.0 to 1.0.
"""
invoice_score = _count_matches(text, _INVOICE_KEYWORDS)
receipt_score = _count_matches(text, _RECEIPT_KEYWORDS)
po_score = _count_matches(text, _PO_KEYWORDS)
scores = {
"invoice": invoice_score,
"receipt": receipt_score,
"purchase_order": po_score,
}
max_score = max(scores.values())
if max_score == 0:
return "unknown", 0.0
best = max(scores, key=scores.get)
total = sum(scores.values())
confidence = max_score / total if total > 0 else 0.0
if max_score >= 3:
confidence = min(0.95, confidence + 0.3)
elif max_score >= 2:
confidence = min(0.85, confidence + 0.15)
elif max_score == 1:
confidence = min(0.65, confidence)
return best, round(confidence, 4)